Early fault diagnosis of bearing based on frequency band extraction and improved tunable Q-factor wavelet transform

Abstract

Early fault signals of bearings feature a wide frequency band and strong background noise, which contain both an oscillation component and a periodic impact component. Given its excellent signal decomposition performance, an improved tunable Q-factor wavelet transform (ITQWT) method is applied to separate a measured bearing fault signal into a high-Q-factor oscillation component and a low-Q-factor periodic impact component. In ITQWT, the Q-factors of TQWT are selected by particle swarm optimization to ensure the accuracy of its decomposition. Due to the weak fault symptoms and heavy noise of early fault signals, ITQWT is difficult to apply directly for early fault detection. Thus, a strategy based on frequency band extraction and ITQWT for early fault diagnosis in bearings is proposed. Frequency band extraction based on frequency slice wavelet transform (FSWT) shows significant advantage on observed frequency band selection, which could be highly suitable to extract early fault transient signals within a resonant frequency band. The improved performance of the proposed strategy is tested in simulated and experimental signals. Relative to other methods such as: TQWT and two spectral kurtosis-based methods, the proposed strategy is suitable and reliable for the weak fault feature detection of bearings at the early fault stage.